opensource/PaddleOCR
3
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleOCR.git synced 2026-09-24 23:33:08 +08:00
Code Issues Packages Projects Releases Wiki Activity

upload rec pipeline

Browse Source
This commit is contained in:
tink2123
2022-05-06 06:21:29 +00:00
parent f179144cfd
commit 5c7146c04d
2 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
doc/doc_ch/PP-OCRv3_introduction.md
+3 -1
View File
@@ -128,7 +128,9 @@ PP-OCRv3 期望在提升模型精度的同时，不带来额外的推理耗时
<img src="../ppocr_v3/GTC.png" width=800>
在蒸馏策略方面，PP-OCRv3参考 [SSL](https://github.com/ku21fan/STR-Fewer-Labels) 设计了文本方向任务，训练了适用于文本识别的预训练模型，加速模型收敛过程，精度提升了0.6%; 使用UDML蒸馏策略，进一步提升精度1.5%，训练流程所示：
在蒸馏策略方面:
PP-OCRv3参考 [SSL](https://github.com/ku21fan/STR-Fewer-Labels) 设计了文本方向任务，训练了适用于文本识别的预训练模型，加速模型收敛过程，精度提升了0.6%; 使用UDML蒸馏策略，进一步提升精度1.5%，训练流程所示：
<img src="../ppocr_v3/SSL.png" width="300"> <img src="../ppocr_v3/UDML.png" width="500">
doc/ppocr_v3/v3_rec_pipeline.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 971 KiB